Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PowerDAG: Reliable Agentic AI System for Automating Distribution Grid Analysis

Emmanuel O. Badmus, Amritanshu Pandey

Abstract

This paper introduces PowerDAG, an agentic AI system for automating complex distribution-grid analysis. We address the reliability challenges of state-of-the-art agentic systems in automating complex engineering workflows by introducing two innovative active mechanisms: (i) \textbf{adaptive retrieval}, which uses a similarity-decay cutoff algorithm to dynamically select the most relevant annotated exemplars as context, and (ii) \textbf{just-in-time (JIT) supervision}, which actively intercepts and corrects tool-usage violations during execution. On a benchmark of unseen distribution grid analysis queries, PowerDAG achieves a 100\% success rate with GPT-5.2 and 94.4--96.7\% with smaller open-source models, outperforming base ReAct (41--88\%), LangChain (30--90\%), and CrewAI (9--41\%) baselines by margins of 6--50 percentage points.

PowerDAG: Reliable Agentic AI System for Automating Distribution Grid Analysis

Abstract

This paper introduces PowerDAG, an agentic AI system for automating complex distribution-grid analysis. We address the reliability challenges of state-of-the-art agentic systems in automating complex engineering workflows by introducing two innovative active mechanisms: (i) \textbf{adaptive retrieval}, which uses a similarity-decay cutoff algorithm to dynamically select the most relevant annotated exemplars as context, and (ii) \textbf{just-in-time (JIT) supervision}, which actively intercepts and corrects tool-usage violations during execution. On a benchmark of unseen distribution grid analysis queries, PowerDAG achieves a 100\% success rate with GPT-5.2 and 94.4--96.7\% with smaller open-source models, outperforming base ReAct (41--88\%), LangChain (30--90\%), and CrewAI (9--41\%) baselines by margins of 6--50 percentage points.
Paper Structure (38 sections, 16 equations, 4 figures, 2 tables, 1 algorithm)

This paper contains 38 sections, 16 equations, 4 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. LLMs for Tool Orchestration
  4. Exemplar-Conditioned Orchestration
  5. Retrieval of Relevant Exemplars
  6. Runtime Supervision of Tool Execution
  7. Preliminaries
  8. Agentic AI System
  9. State Transition and Tool Classification
  10. Workflow as a Directed Acyclic Graph (DAG)
  11. Annotated Query-Workflow Pair Exemplars
  12. An Agentic AI System: PowerDAG
  13. PowerDAG System Architecture
  14. Adaptive Retrieval of Annotated Exemplars
  15. Just-In-Time Supervision
  16. ...and 23 more sections

Figures (4)

  • Figure 1: Workflow as a directed acyclic graph (DAG). Nodes denote tool invocations; directed edges encode ordering-based dependencies.
  • Figure 2: PowerDAG execution architecture. Initialization: the schema extractor builds $\Sigma$ from the tool set, and a retrieval module selects relevant $\mathcal{W}_{sub}$ from $\mathcal{W}_{av}$ for the unseen query $q_u$. Iterative loop: the agent proposes an action, the supervisor either blocks and returns an advisory or executes the action. If executed, an observation returns. The loop stops when the agent gives a textual final response.
  • Figure 3: Two-stage exemplar selection. The system embeds $q_u$ and $\{q_i\}_{i=1}^N$, ranks by cosine similarity, selects an adaptive cutoff $i_\text{cutoff}$, and applies an LLM-based workflow filter to produce the final subset $\mathcal{W}_\text{sub}$.
  • Figure 4: Combined performance score (Pass@1 $\times$ Precision) across models. This metric captures both first-attempt success and workflow correctness. PowerDAG achieves the highest scores on all six models, with near-perfect performance on four of them. The gap between PowerDAG and baselines is largest on smaller models.